International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

EpiGraph: Anomaly Detection in Contact Networks for Early Disease Outbreak Prediction

Authors :

Abolfazl Zarghani¹

1- Ferdowsi university of mashhad

Keywords :

Temporal Graph Neural Networks،Anomaly Detection،Disease Outbreak Prediction،Contact Networks،Epidemiology

Abstract :

Infectious diseases, such as COVID-19, pose severe global health risks, necessitating early detection to enable timely interventions like quarantines and vaccinations. We propose EpiGraph, a neuro-epidemiological framework for detecting anomalies in temporal contact networks to predict disease outbreaks early. EpiGraph integrates Temporal Graph Neural Networks (TGNNs), combining Graph Attention Networks (GATs) for structural learning and Long Short-Term Memory (LSTM) networks for temporal modeling, to capture dynamic human interactions. Epidemiological knowledge prioritizes high-risk contact patterns, such as hospital interactions or those involving high-centrality nodes, enhancing anomaly detection sensitivity. Detected anomalies feed into a hybrid prediction model, merging Transformer-based time series forecasting with dynamically adjusted Susceptible-Infected-Recovered (SIR) models, to generate actionable early warnings. Evaluated on the SocioPatterns dataset, EpiGraph achieves an F1-score of 0.83 for anomaly detection and an AUC of 0.921 for disease prediction, outperforming baselines like StrGNN, STADN, and MPSTAN. Key contributions include unified anomaly detection and prediction, epidemiologically informed modeling, actionable insights via attention mechanisms, and scalability for real-world surveillance. Despite reliance on high-resolution data and simulated anomalies, EpiGraph offers a robust, interpretable solution for public health, with potential to guide targeted interventions in epidemic scenarios.